Mainland China is getting first dibs on a British GPU stack that borrows heavily from Nvidia’s PC playbook. On Tuesday, Sept. 8, 2026, the Xiaomi 18 Fold launches with an Arm Mali G2-Ultra NX graphics processor inside Xiaomi’s custom Xring O3 chip — and, after five years of work, Arm finally has an AI gaming accelerator living inside its mobile GPU.

In a The Verge exclusive by Sean Hollister, Arm frames the Mali G2-Ultra NX two ways. On raw silicon, the company claims about an 85% GPU performance jump versus the GPU Xiaomi was using previously, and roughly 14% over Arm’s prior flagship in “non-AI gaming.” The bigger pitch is the feature set Arm is branding neural graphics: hardware-accelerated AI upscaling (Neural Super Sampling), frame generation (Neural Frame Rate Upscaling), and Neural Denoising — the same family of tricks Nvidia ships as DLSS / ray reconstruction, ported into a phone power envelope.

“You will be able to buy it; you will be able to get it in phones next year, you will be able to get it in other types of larger screen devices, whether that be tablets or Chromebooks,” Arm EVP of Edge AI Chris Bergey told The Verge.

Bergey would not name other SKUs that will ship the GPU, but he hinted some of Google’s forthcoming “Googlebooks” could use the same graphics stack. “We think it’s going to be huge,” he said.

DLSS for Android — with the usual caveats

Arm’s marketing slide puts up to 4× frame rate on the table when Neural Super Sampling, Neural Frame Rate Upscaling, and Neural Denoising run together — because each technique lets the GPU render fewer real pixels and invent the rest. Games will not necessarily feel four times faster. Half of that headline boost is interpolated frames, and frame generation on Nvidia, AMD, and Intel typically adds latency. Treat the 4× chart as a stacking claim, not a promise that Infinity Nikki suddenly plays like a wired console.

“This really is about bringing Tensor-like capabilities into the GPU shader,” Bergey told The Verge.

The rollout is staged. “The first titles will largely be super sampling,” Bergey said, with frame generation arriving in early 2027 and denoising “soon after that.” Sustained power targets are phone-realistic: an entire handset at ≤2.5 W, with about 1.5 W budgeted to the GPU — “basically what you want for real sustained gaming where it’s not getting warm in your phone,” Bergey said. Or partners can underclock for battery. “This is not overclocking, this is mainstream,” he added. “This is about AI replacing traditional rendering; anywhere you can use it, you will use it.”

On the larger Xiaomi Pad 9 Pro Max — same Xring O3, more battery and cooling — Xiaomi is showing 3.2K at 120 fps. Arm also says neural graphics will spill beyond flagship Mali G2-Ultra NX into Premium and Pro tiers later.

Games, engines, and Switch 2 context

Developers still have to integrate the features; Unreal Engine and Unity can help, but Bergey does not expect a universal OS toggle that magically upgrades every sideloaded APK. The first confirmed titles are Where Winds Meet, Infinity Nikki, and Arena Breakout: Infinite, with more “coming weeks” announcements he is not yet allowed to name.

Mobile is not alone in this arms race. Stellar Blade’s Switch 2 build is chasing 60 fps with Nvidia DLSS plus an AMD-style frame-gen path on Nintendo’s Arm-based SoC that carries a custom Nvidia GPU — different silicon, same idea: AI fill-in to stretch limited mobile rasterization.

For now the concrete shipping fact is narrow and dated Tuesday, Sept. 8, 2026: China gets a foldable with Arm’s first AI-native Mali flagship, neural super sampling in early games, frame gen later, and a Verge-sourced roadmap that puts the same stack on phones, tablets, and Chromebooks next year. Whether “4×” survives outside Arm’s slide deck is a question for hands-on frame-time charts — not launch-day marketing.

Sources